DarthAffe/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2025-12-13 05:48:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

apply jeffbolz f32 patch

Browse Source 
https://github.com/leejet/stable-diffusion.cpp/pull/851#issuecomment-3335515302
This commit is contained in:
leejet 2025-10-11 01:04:14 +08:00
parent d19d4a5903
commit 6ea2a75929
3 changed files with 16 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
common.hpp
View File

@ -242,7 +242,8 @@ public:
FeedForward(int64_t dim, FeedForward(int64_t dim,
int64_t dim_out, int64_t dim_out,
int64_t mult = 4, int64_t mult = 4,
Activation activation = Activation::GEGLU) { Activation activation = Activation::GEGLU,
bool force_prec_f32 = false) {
int64_t inner_dim = dim * mult; int64_t inner_dim = dim * mult;
if (activation == Activation::GELU) { if (activation == Activation::GELU) {
@ -252,7 +253,7 @@ public:
} }
// net_1 is nn.Dropout(), skip for inference // net_1 is nn.Dropout(), skip for inference
blocks["net.2"] = std::shared_ptr<GGMLBlock>(new Linear(inner_dim, dim_out)); blocks["net.2"] = std::shared_ptr<GGMLBlock>(new Linear(inner_dim, dim_out, true, false, force_prec_f32));
} }
struct ggml_tensor* forward(struct ggml_context* ctx, struct ggml_tensor* x) { struct ggml_tensor* forward(struct ggml_context* ctx, struct ggml_tensor* x) {

17
ggml_extend.hpp
View File

@ -939,8 +939,12 @@ __STATIC_INLINE__ struct ggml_tensor* ggml_group_norm_32(struct ggml_context* ct
__STATIC_INLINE__ struct ggml_tensor* ggml_nn_linear(struct ggml_context* ctx, __STATIC_INLINE__ struct ggml_tensor* ggml_nn_linear(struct ggml_context* ctx,
struct ggml_tensor* x, struct ggml_tensor* x,
struct ggml_tensor* w, struct ggml_tensor* w,
struct ggml_tensor* b) { struct ggml_tensor* b,
bool force_prec_f32 = false) {
x = ggml_mul_mat(ctx, w, x); x = ggml_mul_mat(ctx, w, x);
if (force_prec_f32) {
ggml_mul_mat_set_prec(x, GGML_PREC_F32);
}
if (b != NULL) { if (b != NULL) {
x = ggml_add_inplace(ctx, x, b); x = ggml_add_inplace(ctx, x, b);
} }
@ -1953,6 +1957,7 @@ protected:
int64_t out_features; int64_t out_features;
bool bias; bool bias;
bool force_f32; bool force_f32;
bool force_prec_f32;
void init_params(struct ggml_context* ctx, const String2GGMLType& tensor_types = {}, const std::string prefix = "") { void init_params(struct ggml_context* ctx, const String2GGMLType& tensor_types = {}, const std::string prefix = "") {
enum ggml_type wtype = get_type(prefix + "weight", tensor_types, GGML_TYPE_F32); enum ggml_type wtype = get_type(prefix + "weight", tensor_types, GGML_TYPE_F32);
@ -1969,12 +1974,14 @@ protected:
public: public:
Linear(int64_t in_features, Linear(int64_t in_features,
int64_t out_features, int64_t out_features,
bool bias = true, bool bias = true,
bool force_f32 = false) bool force_f32 = false,
bool force_prec_f32 = false)
: in_features(in_features), : in_features(in_features),
out_features(out_features), out_features(out_features),
bias(bias), bias(bias),
force_f32(force_f32) {} force_f32(force_f32),
force_prec_f32(force_prec_f32) {}
struct ggml_tensor* forward(struct ggml_context* ctx, struct ggml_tensor* x) { struct ggml_tensor* forward(struct ggml_context* ctx, struct ggml_tensor* x) {
struct ggml_tensor* w = params["weight"]; struct ggml_tensor* w = params["weight"];
@ -1982,7 +1989,7 @@ public:
if (bias) { if (bias) {
b = params["bias"]; b = params["bias"];
} }
return ggml_nn_linear(ctx, x, w, b); return ggml_nn_linear(ctx, x, w, b, force_prec_f32);
} }
}; };

2
qwen_image.hpp
View File

@ -196,7 +196,7 @@ namespace Qwen {
blocks["img_norm1"] = std::shared_ptr<GGMLBlock>(new LayerNorm(dim, eps, false)); blocks["img_norm1"] = std::shared_ptr<GGMLBlock>(new LayerNorm(dim, eps, false));
blocks["img_norm2"] = std::shared_ptr<GGMLBlock>(new LayerNorm(dim, eps, false)); blocks["img_norm2"] = std::shared_ptr<GGMLBlock>(new LayerNorm(dim, eps, false));
blocks["img_mlp"] = std::shared_ptr<GGMLBlock>(new FeedForward(dim, dim, 4, FeedForward::Activation::GELU)); blocks["img_mlp"] = std::shared_ptr<GGMLBlock>(new FeedForward(dim, dim, 4, FeedForward::Activation::GELU, true));
// txt_mod.0 is nn.SiLU() // txt_mod.0 is nn.SiLU()
blocks["txt_mod.1"] = std::shared_ptr<GGMLBlock>(new Linear(dim, 6 * dim, true)); blocks["txt_mod.1"] = std::shared_ptr<GGMLBlock>(new Linear(dim, 6 * dim, true));